Extended Transform Partitions for Video Compression

1. A method for encoding a video block, the method comprising: performing prediction-level partitioning against the video block to partition the video block into one or more first sub-blocks; generating a prediction residual block for a first sub-block of the one or more first sub-blocks based on a prediction of the first sub-block; determining a transform block partition type for performing transform-level partitioning against the prediction residual block, wherein the transform block partition type is one of multiple transform block partition types available for partitioning the prediction residual block; limiting transform-level partitioning performance to a single, non-recursive transform-level partitioning against the prediction residual block using the transform block partition type to partition the prediction residual block into a plurality of second sub-blocks; generating a transform block for a second sub-block of the plurality of second sub-blocks by transforming video data of the second sub-block; and encoding the transform block and a symbol representative of the transform block partition type to a bitstream.

2. The method of claim 1 , wherein the prediction residual block is a first prediction residual block, the transform block partition type is a first transform block partition type, and the transform block is a first transform block, and wherein the method further comprises: generating a second prediction residual block for another first sub-block of the one or more first sub-blocks based on a prediction of the other first sub-block; determining a second transform block partition type for performing transform-level partitioning against the second prediction residual block, wherein the second transform block partition type is one of the multiple transform block partition types and is different than the first transform block partition type; performing a single, non-recursive transform-level partitioning against the second prediction residual block using the second transform block partition type to partition the second prediction residual block into a plurality of other second sub-blocks; generating a second transform block for another second sub-block of the plurality of other second sub-blocks by transforming video data of the other second sub-block; and encoding the second transform block and a symbol representative of the second transform block partition type to a bitstream, wherein the symbol representative of the second transform block partition type is different than the symbol representative of the first transform block partition type.

3. The method of claim 1 , wherein a shape of each second sub-block of the plurality of second sub-blocks is different than a shape of the prediction residual block.

4. The method of claim 1 , wherein determining the transform block partition type for performing the transform-level partitioning against the prediction residual block comprises: using a learning model to determine, based on the prediction residual block, to exclude one or more transform block partition types of the multiple transform block partition types from evaluation, wherein the transform block partition type is other than one of the excluded one or more transform block partition types.

5. The method of claim 1 , wherein the symbol representative of the transform block partition type is an entropy coded symbol defined to uniquely represent the transform block partition type.

6. The method of claim 1 , wherein the transform block generated for the second sub-block has a size equal to one of 4×64, 64×4, 4×32, 32×4, 8×64, or 64×8.

7. The method of claim 1 , wherein the plurality of second sub-blocks includes up to four second sub-blocks, wherein each of the up to four second sub-blocks corresponds to at least one fourth of video data of the prediction residual block.

8. The method of claim 1 , wherein one or more transform block partition types of the multiple transform block partition types are excluded from evaluation based on a size of the prediction residual block.

9. A method for decoding an encoded video block from a bitstream, the method comprising: deriving a first symbol and a second symbol from the bitstream, the first symbol representing a first transform block partition type used during a transform-level partitioning performance limited to a single, non-recursive transform-level partitioning of a first prediction residual block including first video data of the encoded video block, the second symbol representing a second transform block partition type used during a transform-level partitioning performance limited to a single, non-recursive transform-level partitioning of a second prediction residual block including second video data of the encoded video block, wherein the first transform block partition type and the second transform block partition type are different; producing the first prediction residual block according to the first symbol by inverse transforming and combining a plurality of first encoded transform sub-blocks associated with the first video data; producing the second prediction residual block according to the second symbol by inverse transforming and combining a plurality of second encoded transform sub-blocks associated with the second video data; reconstructing a video block including the first video data and the second video data using the first prediction residual block and the second prediction residual block; and outputting the video block within an output video stream for storage or display.

10. The method of claim 9 , wherein a selection of the first transform block partition type for the first prediction residual block during an encoding of the encoded video block and a selection of the second transform block partition type for the second prediction residual block during the encoding of the encoded video block are both independent of whether the first video data and the second video data are encoded using an intra-prediction mode or an inter-prediction mode.

11. The method of claim 9 , wherein a shape of first inverse transformed blocks produced by inverse transforming each of the plurality of first encoded transform sub-blocks is different than a shape of the first prediction residual block, and wherein a shape of second inverse transformed blocks produced by inverse transforming each of the plurality of second encoded transform sub-blocks is different than a shape of the second prediction residual block.

12. The method of claim 9 , wherein the first transform block partition type and the second transform block partition type are each selected from multiple available transform block partition types, wherein a learning model is used during the encoding of the encoded video block to determine ones of the multiple available transform block partition types to exclude one or both of a selection of the first transform block partition type for the first prediction residual block or a selection of the second transform block partition type for the second prediction residual block.

13. The method of claim 9 , wherein each of the plurality of first encoded transform sub-blocks and each of the plurality of second encoded transform sub-blocks has a size equal to one of 4×64, 64×4, 4×32, 32×4, 8×64, or 64×8.

14. The method of claim 9 , wherein each of the plurality of first encoded transform sub-blocks corresponds to at least one fourth of the first video data and each of the plurality of second encoded transform sub-blocks corresponds to at least one fourth of the second video data.

15. A method for decoding an encoded video block from a bitstream, the method comprising: deriving, from a bitstream to which the encoded video block is encoded, a symbol representing a transform block partition type used during a transform-level partitioning performance limited to a single, non-recursive transform-level partitioning of a prediction residual block encoded to produce the encoded video block; producing a first inverse transformed block by inverse transforming a first encoded transform sub-block associated with first video data of the prediction residual block; producing a second inverse transformed block by inverse transforming a second encoded transform sub-block associated with second video data of the prediction residual block; producing, according to the transform block partition type represented by the symbol, the prediction residual block using the first inverse transformed block and the second inverse transformed block; reconstructing a video block including the first video data and the second video data using the prediction residual block; and outputting the video block within an output video stream for storage or display.

16. The method of claim 15 , wherein the symbol is a first symbol, the transform block partition type is a first transform block partition type, and the prediction residual block is a first prediction residual block, and wherein the method further comprises: deriving, from the bitstream, a second symbol representing a second transform block partition type used for non-recursive transform-level partitioning of a second prediction residual block associated with the encoded video block; producing a third inverse transformed block by inverse transforming a third encoded transform sub-block associated with third video data of the second prediction residual block; producing a fourth inverse transformed block by inverse transforming a fourth encoded transform sub-block associated with fourth video data of the second prediction residual block; and producing, according to the second transform block partition type represented by the symbol, the second prediction residual block using the third inverse transformed block the fourth inverse transformed block, and wherein reconstructing the video block including the first video data and the second video data using the prediction residual block comprises reconstructing the video block using the first prediction residual block and the second prediction residual block.

17. The method of claim 15 , wherein a shape of each of the first inverse transformed block and the second inverse transformed block is different than a shape of the prediction residual block.

18. The method of claim 15 , wherein the transform block partition type is selected from the at least some of multiple transform block partition types, wherein a learning model is used during the encoding of the encoded video block to determine ones of the multiple transform block partition types to exclude from a selection of the transform block partition type for the prediction residual block.

19. The method of claim 15 , wherein each of the first inverse transformed blocks and the second inverse transformed block has a size equal to one of 4×64, 64×4, 4×32, 32×4, 8×64, or 64×8.

20. The method of claim 15 , wherein the first inverse transformed block corresponds to at least one fourth of the first video data and the second inverse transformed block corresponds to at least one fourth of the second video data.